Bearing fault diagnosis based on STFT-SWT and two-flow CNN-KNN

To address the issue that traditional bearing fault diagnosis relies heavily on expert system, and its shallow machine learning method is not effective for bearing fault identification. This paper introduces a model for bearing fault diagnosis utilizing STFT-SWT and two-stream CNN-KNN.To begin with, the vibration signals from non-stationary bearings are transformed into high-frequency time-frequency representations, creating a two-dimensional depiction. These representations are then utilized by a Convolutional Neural Network (CNN) to achieve accurate bearing fault diagnosis. Finally, the fault diagnosis classification results are output by KNN. In order to solve the problem of uneven time-frequency resolution of short-time Fourier transform, it is combined with synchronous compressed wavelet transform. Experimental results using the bearing dataset from Case Western Reserve University indicate that the diagnostic performance of this method surpasses that of other traditional models and effectively enhances the diagnostic accuracy of bearings.